Demographic effects on the composition of children’s early vocabulary

As vocabulary grows…

…children vary immensely

with subgroups of variation

Decades of research has used parent report vocabulary checklists – the CDI – to quantify and <> language development. A large body of research using this and other methods has revealed that girls tend to know more words than boys, that earlier-born children tend to know more words than later-born children, and that children from higher SES families tend to more words than children from lower-SES families. This last set of findings has received particular attention – as we’ve seen from the other presentations in this symposium, it has important real-world consequences for children’s lives. Grouped by SES, the quantity and quality of language input that children receive predicts their vocabulary size and other language skills, and vocabulary at only 25 months old predicts cognitive and linguistic abilities at 8 years old.

How do we go beyond a simple breakdown of vocabulary by demographic factor and beyond just a correlation between input and vocabulary? To ask more fine-grained questions, we need more data…

median productive vocabulary at 24 months in english

girls 368.7, boys 272.4 (1.4 times larger) [wordbank]

first 355.3, second 297.6 (1.2 times larger) [wordbank]

higher SES 441.5, lower 287.9 (1.5 times larger) [fernald2013]

lowest SES group less than a third of words of highest SES group

Vocabulary size

Girls > boys

Eriksson et al, 2012; Frank et al, 2016
First-born > later-born

Goldfield & Reznick, 1990
Higher SES > lower SES

Arriaga et al 1998; Fernald, Marchman, & Weisleder, 2013

Quantity/quality of language input → vocabulary size

Hart & Risley, 1995; Hoff, 2003
Vocabulary at 25 months → abilities at 8 years

Marchman & Fernald, 2008

Do individual words carry demographic signal?

Data

Sex
– Assigned female at birth
– Assigned male at birth

Data

Birth order
– First-born
– Second-born
– Third-born or later-born

Data

Maternal education
– Below Secondary
– Secondary (~high school degree)
– College and Above (~college degree)

Analysis

For a language and demographic:

produces ~ age + birth_order + (age + birth_order | item)

fixed effect for each demographic in each language
random effect for each item for each demographic in each language
coding scheme gives difference between adjacent levels

Results

Item results

Takeaways

Across languages, demographic factors have consistent direction but variable magnitude
Individual words are learned differentially by sex, birth order, maternal education
Avoid bias in test design by excluding demographically-linked items
Demographic differences in vocabulary size are influenced by specific content differences

Part of larger work on variability and consistency in language learning:
langcog.github.io/wordbank-book

All data and code available at: github.com/mikabr/item-demo